The field of pulsar candidate identification still faces the challenge of algorithm generalization, as a single model often fails to adapt to datasets with diverse sources and characteristics. To address this issue, we propose a Genetic Algorithm for Multi-Modal Adaptive Convolutional Neural Network (GAMMA-CNN), which identifies pulsar candidates using diagnostic-style features derived from pulsar search pipelines. This model employs a genetic algorithm to automatically construct network architectures, introducing skip connections and multimodal fusion mechanisms. It can adaptively explore architectures under different modality configurations, thereby obtaining structures well suited to the current observational mode. GAMMA-CNN operates on diagnostic-style features and incorporates a flexible interface that automatically detects 1D and 2D plot formats, enabling convenient dimensional alignment and multimodal fusion. To evaluate the performance of GAMMA-CNN, we designed a series of experiments on the High Time Resolution Universe (HTRU) and Five-hundred-meter Aperture Spherical Telescope (FAST) datasets, covering both unimodal and multimodal inputs, including profile, dispersion measure curve (DM-curve), subband, subintegration, period-dispersion measure (HTRU only), and diagnostic plots. Results show that multimodal fusion enables the network to capture pulsar signal features more comprehensively and surpass the performance limits of unimodal models. When trained with multimodal data, comprising profile, DM-curve, subband, and subintegration plots, GAMMA-CNN achieved an F1 score of 97.79%, recall of 95.80%, and precision of 99.80% on the HTRU dataset and an F1 score of 99.70%, recall of 99.40%, and precision of 100.0% on the FAST dataset, demonstrating its performance across multiple modality settings on the HTRU and FAST datasets.

This study addresses the need for real-time sampling and transmission of wideband signals in radio astronomy, developing a comprehensive technical workflow from hardware implementation to firmware compilation. The system, based on the Xilinx ZCU111 Radio Frequency System on Chip (RFSoC) development board, incorporates hardware modifications and custom development using the CASPER open-source toolflow. We adopt a channelization method based on the Polyphase Filter Bank algorithm, provide the corresponding theoretical derivations, and implement the scheme on the CASPER open-source toolflow. The developed firmware enables channelization (16 channels, 128 MHz bandwidth) of ultra-wideband signals at a 4096 MHz sampling rate on RFSoC hardware. Deployed at the Nanshan 26 m radio telescope for L-band (964-1732 MHz) observations, the system successfully produced high signal-to-noise ratio pulsar profiles after processing with DSPSR software, validating the reliability of the channelization algorithm. These results confirm the feasibility of RFSoC-based architectures for wideband signal channelization, offering a practical pathway for future astronomical backends.

Owing to its large aperture and advanced receivers,research plans for the Qitai 110 m radio telescope(QTT)include a variety of spectral line scientific studies.Sequential construction of receiver systems and multidisciplinary planning require overcoming serious challenges to spectral line digital backend development,notably to digitize,process,and transmit considerable quantities of observational data,to minimize time-to-science with an easily scalable architecture,and to provide robust,high-quality data.As a proof-of-concept for the QTT backend,here we implement a baseband spectral line digital backend with a SNAP+GPU architecture.The SNAP-based digital frontend comprises two digitization links(1 000 MHz,8-bit),two parallel quad-channel preprocessing modules,a quantization module,and a finite-state packaging module,generating a 100-MHz bandwidth digital link from the original analog signal through high-speed Ethernet transmission.The GPU node receives preprocessed baseband packets,constructs a ring buffer for lossless unpacking and distributing,with real-time data reception and caching,and conducts real-time spectral analysis(frequency resolution:3.051 kHz)of the 100 MHz baseband data.We evaluated system performance experimentally using spectral line observations with the Nanshan 26-m radio telescope(NSRT).For the QTT,the SNAP digital frontend will be seamlessly migrated to a radio frequency system-on-chip(RFSoC)architecture,resulting in five-and tenfold increases in instantaneous bandwidth and data throughput,respectively.The low-coupling digital frontend and GPU node can be easily extended to multiple nodes.

The open-source Collaboration for Astronomy Signal Processing and Electronics Research(CASPER)toolflow has become a popular choice for building reconfigurable digital backends in radio astronomy.We extend this toolflow to the third-party TQ47DR Radio-Frequency System-on-Chip(RFSoC)platform,a cost-effective and widely available board.Our implementation includes a custom PetaLinux system,a lightweight control server,a deterministic clock-management driver,and platform-specific yellow-block adaptations that expose the on-chip data converters and 100-Gigabit Ethernet interfaces.System validation demonstrates high-quality converter performance,stable packet streaming,and real-time spectrometry,confirming that such third-party hardware can be integrated seamlessly into this open-source programming ecosystem for next-generation instruments.

Accurate multichannel phase calibration is essential for digital beamforming in next-generation radio telescopes and phased-array feeds (PAFs). This article presents a dual radio frequency system-on-chip (RFSoC)-based multichannel phase-measurement and calibration system that combines gigahertz-class sampling, on-board polyphase filter-bank (PFB) channelization, and frequency-domain cross-spectral processing to estimate relative phase and gain offsets and to derive complex calibration weights for channel equalization and beamforming. A 1.25-GHz probe tone, sampled at 2.048 GS/s, is aliased into the 704-960 MHz band and used as a controlled calibration reference. The proposed methodology is validated in three stages. First, a two-channel phase-shifter experiment quantifies the intrinsic accuracy of the cross-spectral estimator, i.e., the phase estimator derived from the complex cross spectrum between each channel and a reference channel, as a function of signal-to-noise ratio (SNR) and integration length, showing that the root-mean-square (rms) phase error follows the expected 1/\root N behavior and can be reduced below 0.02 degrees over the tested SNR range. Second, an eight-channel coaxial-split experiment demonstrates static multichannel equalization, achieving better than 0.1 degrees phase matching and 0.03-dB gain matching, limited mainly by thermal noise. Third, the resulting calibration weights are applied to a horn-illuminated 4 & times; 4 microstrip array at 1.25 GHz, where the measured digital beams show lower sidelobe levels and improved integrated sidelobe ratios relative to the uncalibrated case, while preserving the main-lobe width and pointing within the measurement resolution. These results indicate that RFSoC-based front ends combined with cross-spectral calibration provide a practical experimental platform for multichannel phased-array measurement and beam-pattern characterization under controlled laboratory conditions.

The most significant distinction between Phased Array Feed(PAF) and traditional multi-beam receivers lies in its beamforming capability, where the purpose of beam calibration is to accurately determine the weighting information for each beam within the field of view. This paper first introduces the beamforming principles and algorithm classifications of PAF. Drawing on the successfully implemented PAF beam calibration schemes of Australian Square Kilometer Array Pathfinder and Aperture Tile in Focus in radio astronomy observations, we present a detailed explanation of the principles and procedures for beam calibration based on the Maximum Signal-to-Noise Ratio(Max-SNR) and Linear Constrained Minimum Variance(LCMV) algorithm. Based on the existing 1.25 GHz microstrip antenna PAF array, we designed a calibration grid comprising seven distinct directional beams and established an experimental platform for PAF array beam calibration based on an analog beamformer with the Max-SNR algorithm. Test results demonstrated a maximum gain fluctuation of 3.2 d B among the seven beams, and we further refined the weighting coefficients for each beam by selecting an appropriate gain target value and introducing a scaling factor, thereby reducing the maximum gain fluctuation between the seven beams to 1.5 dB. Building upon the Max-SNR measurements of the central axial beam's intersection points with adjacent beam patterns and their gain differentials, we implemented a recalibration of the central beam using the LCMV algorithm. This approach enforced gain consistency across predetermined directions, with test results demonstrating a reduction in the central beam's gain fluctuation from 1.6 to 0.9 dB.This research systematically validates the array-level beamforming algorithm and beam calibration scheme,offering significant guidance for future PAF receivers in areas such as beam calibration scheme selection and performance correction. Furthermore, it establishes a solid foundation for the development of advanced digital beamforming technologies with enhanced data processing ability and weighting accuracy.

This paper addresses the performance degradation issue in a fast radio burst search pipeline based on deep learning.This issue is caused by the class imbalance of the radio frequency interference samples in the training dataset,and one solution is applied to improve the distribution of the training data by augmenting minority class samples using a deep convolutional generative adversarial network.Experimental results demonstrate that retraining the deep learning model with the newly generated dataset leads to a new fast radio burst classifier,which effectively reduces false positives caused by periodic wide-band impulsive radio frequency interference,thereby enhancing the performance of the search pipeline.

This paper achieves indirect monitoring and prediction of the working conditions of the refrigeration system by monitoring the temperature of the receiver Dewar, thereby ensuring that the radio telescope can maintain its optimal operating state continuously and thus guarantee its observational sensitivity. This goal is achieved by fully utilizing existing temperature monitoring data, without the need for additional hardware investment, thus reducing costs and enhancing practicality. By setting thresholds and identification factors, this paper employs the K-means++ clustering algorithm to process the temperature data of the receiver Dewar. The algorithm classifies the data into four categories, representing normal, temperature rise, temperature drop, and ambient temperature states. The results indicate that by monitoring the temperature characteristics of the receiver Dewar, this paper can conduct in-depth analysis of key information such as the refrigeration conditions of the receiver and the gas tightness of the Dewar container, thereby providing a solid basis for fault prediction and maintenance. Furthermore, the K-means++ algorithm exhibits good performance and applicability when processing such data, providing strong support for subsequent research and applications in this paper. This research finding is of great significance for improving the observational sensitivity and stability of radio telescopes.

Pulsar candidate identification is an indispensable task in pulsar science.Based on the characteristics of imbalanced and diverse pulsar data sets,and the lack of a unified processing framework,we first used dimensionality reduction and visualization to analyze potential deficiencies caused by the incompleteness of current data set extraction methods.We found that the limited use of non-pulsar data may lead to bias in the result,which may limit the generalization ability.Based on the dimensionality reduction results,we propose a Grid Group Uniform Sampling(GGUS) method.This data preprocessing method improves the performance of Random Forest,Support Vector Machine,Convolutional Neural Network,and Res Net50 models on Lyon’s features,diagnostic plots,and perioddispersion measure (period-DM) plots in the HTRU1 data set.The average recall increased by approximately0.5%,precision by nearly 2%,and F1score by around 1.2%for all models and in all data sets.In the period-DM plots testing,the high-performance Res Net50 algorithm achieved over 98%F1using random sampling.GGUS demonstrated further improvements in this test,enhancing the average F1score,precision,and recall by approximately 0.07%,0.1%,and 0.03%,respectively.

Aiming at the problem of radio observation data quality degradation or invalid data caused by Radio Frequency Interference (RFI), we have studied the adaptive RFI algorithm and proposed an interference mitigation method based on the adaptive filter. An adaptive interference suppression model has been established to effectively counter complex and intense interference. This model includes a normalized factor, derived from the filter's correlation function, which improves the performance of the adaptive filtering algorithm in reducing RFI. The adaptive filter algorithm flexibly adjusts the filter gain automatically, based on the correlation between the RFI reference signal and the observation signal, thereby effectively eliminating the RFI mixed with astronomical signals. Simulation results show that the algorithm can effectively eliminate radio frequency interference, improve the signal-to-noise ratio of astronomical signals and signal recognition rate. It also prevents the loss of signal bandwidth, ensuring the provision of reliable data for subsequent post-processing analysis.

This paper presents two cryogenic low-noise amplifiers (LNAs) based on the WIN’s 0.18 μm gate length gallium arsenide (GaAs) pseudomorphic high electron mobility transistor (pHEMT) process designed for radio telescope receivers. Discrete transistors with gate peripheries spanning 50–600 μm were DC-characterized at 290 K and 15 K, respectively. The LNAs underwent on-chip noise characterization under 15 K using a Y-factor measurement setup, which integrated a calibrated noise source and a noise figure analyzer. This approach directly quantified the noise temperature—critical metrics for radio telescope receiver front-ends. The top-performing LNA variant identified through on-chip characterization was packaged and evaluated in a cryogenic test-bed. This LNA, spanning a bandwidth of 0.3–15 GHz, demonstrated a gain of 26 dB and a minimum noise temperature of 6 K when operated at an ambient temperature of 15 K. In contrast, a second LNA architecture, tested solely on-chip, demonstrated a gain of 30 dB and a minimum noise temperature of 15 K across the 0.3–7 GHz range.

We introduce the structure of a radio astronomy phased array feeds(PAF)beamforming demonstrator.In a laboratory environment,we have demonstrated beamforming on a received 1.25 GHz sinusoidal signal and used digital weighting techniques to plot the 2D pattern of the PAF.The radio frequency part of the demonstrator includes a 4×4 linearly polarized microstrip antenna array,all of which is connected in series with a low-noise amplifier.The signals from the central 4×2 array elements are injected into a radio frequency system-on-chip digital board,which can receive eight inputs with a bandwidth of 512 MHz.Combining the principle of undersampling,the beamforming is completed at a frequency of 1.25 GHz for the offline data,and a 2D image of the beam is plotted using beam scanning technology.

Fast radio bursts (FRBs) are among the most studied radio transients in astrophysics, but their origin and radiation mechanism are still unknown. It is a challenge to search for FRB events in a huge amount of observational data with high speed and high accuracy. With the rapid advancement of the FRB research process, FRB searching has changed from archive data mining to either long-term monitoring of the repeating FRBs or all-sky surveys with specialized equipments. Therefore, establishing a highly efficient and high quality FRB search pipeline is the primary task in FRB research. Deep learning techniques provide new ideas for FRB search processing. We have detected radio bursts from FRB 20201124A in the L-band observational data of the Nanshan 26 m radio telescope (NSRT-26m) using the constructed deep learning based search pipeline named dispersed dynamic spectra search (DDSS). Afterwards, we further retrained the deep learning model and applied the DDSS framework to S-band observations. In this paper, we present the FRB observation system and search pipeline using the S-band receiver. We carried out search experiments, and successfully detected the radio bursts from the magnetar SGR J1935+2145 and FRB 20220912A. The experimental results show that the search pipeline can complete the search efficiently and output the search results with high accuracy.

With a focus on the requirements of electromagnetic compatibility measurement and electromagnetic protection design of the Qitai 110-m radio telescope, herein, we constructed a 3-m radio anechoic chamber measurement platform, which provided essential data and technical support for electromagnetic compatibility design and measurement -based performance verification of the large radio telescope. First, we elaborated on the composition of the constructed measurement platform, related equipment performance parameters, the configuration links of the components, and the involved measurement methods, which could be used to perform shielding effectiveness, radiation emission, and radiation susceptibility measurements. Subsequently, we analyzed the differences between the testing results of two measurement standards (GJB 151B and GB/T 9254). Based on the obtained results, we outlined practical application considerations and application examples of the measurement platform, providing important technical support for the electromagnetic compatibility design of large radio telescopes.

The receiver is a signal receiving device placed at the focus of the telescope. In order to improve the observation efficiency, the concept of phased array receiver has been proposed in recent years, which places a small phased array at the focal plane of the reflector, and flexible pattern and beam scanning functions can be achieved through a beamforming network. If combined with the element multiplexing, all beams within the entire field of view can be observed simultaneously to achieve continuous sky coverage. This article focuses on the front-end array of phased array receiver at 0.7–1.8 GHz for QiTai Telescope, and designs a Vivaldi antenna array of PCB structure with dual line polarization. Each polarization antenna is designed to arrange in a rectangle manner by 11 × 10. Based on the simulation results of the focal field, 32, 18, and eight elements were selected to form one beam at 0.7, 1.25, and 1.8 GHz. An analog beamforming network was constructed, and the measured gains of axial beam under uniform weighting were 19.32, 13.72, and 15.22 dBi. Combining the beam scanning method of reflector antenna, the pattern test of different position element sets required for PAF beam scanning was carried out under independent array. The pattern optimization at 1.25 GHz was carried out by weighting method of conjugate field matching. Compared with uniform weighting, the gain, sidelobe level, and main beam direction under conjugate field matching have been improved. Although the above test and simulation results are slightly different, which is related to the passive array and laboratory testing condition, the relevant work has accumulated experience in the development of the front-end array for the phased array receiver, and has good guiding significance for future performance verification after the array is installed on the telescope.

The simulation of radio frequency interference (RFI) cancellation by applying a spatial filtering technique for phased array feed (PAF) is presented. In order to better reflect the characteristics of PAF, a new signal model is to add the coupling coefficient among elements of PAF to the conventional array signal model. Then the subspace projection (SP) algorithm is used to cancel RFI from the correlation matrix of the signal, and finally, the 2D power image is drawn. The power variation of signal-of-interest direction and RFI direction before and after using the SP algorithm is analyzed. The new signal model and simulation strategy can be used to test and verify the beamformer.

A phased array feed(PAF)is a type of receiving array that places phased array antennas on the focal plane of a radio telescope to expand its field of view and improve observation efficiency.Owing to the mutual coupling effect between elements caused by a tightly arranged feed array,which changes the performance of a PAF,this paper presents a 7×7 rectangular feed array model for a 25 m reflector telescope.By adjusting the element spacings,the performance of a PAF with different spacings is comprehensively analyzed with respect to the mutual coupling effect via performance statistics and comparison.This research aims to provide a reference for the preliminary design of a related PAF.

To quickly search for rare fast radio bursts (FRBs) from massive astronomical observational data, Radio Frequency interference (RFI) mitigation is one of the key and challenging task. RFI will cause search algorithm to output a large number of false positive candidates, and even submerges real astronomical events. Due to the complexity of the sources and types of RFI. there is currently no universal method to solve this problem. In order to reduce the impact of RFI on FRB search, the RFI in L band observational data of the Nanshan 26m radio telescope (NSRT-26m) was analyzed and studied. A three-layer RFI mitigation procedure was established for the main narrowband RFI and broadband RFI, which effectively alleviated the RFI pollution of observational data. Embedding this procedure into the FRB DDSS (Dispersion Dynamic Spectra Search) pipeline, experimental results show that the detection rate and accuracy of the search algorithm have been further improved. This method provides valuable reference for RFI mitigation of FRB observational data.